在已知图像观测值和相机内外参数的多视图三角化中,由于观测噪声的存在,导致中点法和L2反投影标准法分别在三角化精度和效率上存在不足。因此,提出了一种基于逆深度自适应加权的多视图三角化方法。首先,通过构建待估计空间三维点在多视图环境下的逆深度模型,赋予不同视点下观测误差对应的自适应权重。然后,确定多视图三角化近似角度误差的无偏估计模型。最后,利用固定点迭代快速求解代价函数。在仿真和实际数据集上的实验结果表明,本方法能很好地平衡多视图三角化的精度和效率,且在不同噪声情况下的重建精度和迭代次数有较强的鲁棒性。

In the multi-view triangulation with known image observation values and camera internal and external parameters, due to the existence of observation noise, the midpoint method and the L2 back projection standard method have insufficient triangulation accuracy and efficiency, respectively. Therefore, this paper proposes an inverse depth adaptive weihting based multi-view triangulation method. First, by constructing an inverse depth model of the three-dimensional points to be estimated in a multi-view environment, the corresponding weights to the observation errors are assigned under different viewpoints. Then an unbiased estimation model of the approximate angle error for the multi-view triangulation is determined. Finally, a fixed-point iteration is carried out to quickly solve the cost function. Experimental results both on simulation and real datasets show that the proposed method can obtain a better balance between accuracy and efficiency for multi-view triangulation, and the reconstruction accuracy and the number of iterations under different noise conditions are robust.